kernel/fork.c

   1 /*
   2  *  linux/kernel/fork.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  *  'fork.c' contains the help-routines for the 'fork' system call
   9  * (see also entry.S and others).
  10  * Fork is rather simple, once you get the hang of it, but the memory
  11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  12  */
  13
  14 #include <linux/slab.h>
  15 #include <linux/init.h>
  16 #include <linux/unistd.h>
  17 #include <linux/module.h>
  18 #include <linux/vmalloc.h>
  19 #include <linux/completion.h>
  20 #include <linux/personality.h>
  21 #include <linux/mempolicy.h>
  22 #include <linux/sem.h>
  23 #include <linux/file.h>
  24 #include <linux/fdtable.h>
  25 #include <linux/iocontext.h>
  26 #include <linux/key.h>
  27 #include <linux/binfmts.h>
  28 #include <linux/mman.h>
  29 #include <linux/mmu_notifier.h>
  30 #include <linux/fs.h>
  31 #include <linux/nsproxy.h>
  32 #include <linux/capability.h>
  33 #include <linux/cpu.h>
  34 #include <linux/cgroup.h>
  35 #include <linux/security.h>
  36 #include <linux/hugetlb.h>
  37 #include <linux/swap.h>
  38 #include <linux/syscalls.h>
  39 #include <linux/jiffies.h>
  40 #include <linux/tracehook.h>
  41 #include <linux/futex.h>
  42 #include <linux/compat.h>
  43 #include <linux/task_io_accounting_ops.h>
  44 #include <linux/rcupdate.h>
  45 #include <linux/ptrace.h>
  46 #include <linux/mount.h>
  47 #include <linux/audit.h>
  48 #include <linux/memcontrol.h>
  49 #include <linux/ftrace.h>
  50 #include <linux/profile.h>
  51 #include <linux/rmap.h>
  52 #include <linux/ksm.h>
  53 #include <linux/acct.h>
  54 #include <linux/tsacct_kern.h>
  55 #include <linux/cn_proc.h>
  56 #include <linux/freezer.h>
  57 #include <linux/delayacct.h>
  58 #include <linux/taskstats_kern.h>
  59 #include <linux/random.h>
  60 #include <linux/tty.h>
  61 #include <linux/proc_fs.h>
  62 #include <linux/blkdev.h>
  63 #include <linux/fs_struct.h>
  64 #include <linux/magic.h>
  65 #include <linux/perf_event.h>
  66 #include <linux/posix-timers.h>
  67 #include <linux/user-return-notifier.h>
  68 #include <linux/oom.h>
  69 #include <linux/khugepaged.h>
  70
  71 #include <asm/pgtable.h>
  72 #include <asm/pgalloc.h>
  73 #include <asm/uaccess.h>
  74 #include <asm/mmu_context.h>
  75 #include <asm/cacheflush.h>
  76 #include <asm/tlbflush.h>
  77
  78 #include <trace/events/sched.h>
  79
  80 /*
  81  * Protected counters by write_lock_irq(&tasklist_lock)
  82  */
  83 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  84 int nr_threads;                 /* The idle threads do not count.. */
  85
  86 int max_threads;                /* tunable limit on nr_threads */
  87
  88 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  89
  90 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  91
  92 #ifdef CONFIG_PROVE_RCU
  93 int lockdep_tasklist_lock_is_held(void)
  94 {
  95         return lockdep_is_held(&tasklist_lock);
  96 }
  97 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
  98 #endif /* #ifdef CONFIG_PROVE_RCU */
  99
 100 int nr_processes(void)
 101 {
 102         int cpu;
 103         int total = 0;
 104
 105         for_each_possible_cpu(cpu)
 106                 total += per_cpu(process_counts, cpu);
 107
 108         return total;
 109 }
 110
 111 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 112 # define alloc_task_struct_node(node)           \
 113                 kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node)
 114 # define free_task_struct(tsk)                  \
 115                 kmem_cache_free(task_struct_cachep, (tsk))
 116 static struct kmem_cache *task_struct_cachep;
 117 #endif
 118
 119 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
 120 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
 121 {
 122 #ifdef CONFIG_DEBUG_STACK_USAGE
 123         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
 124 #else
 125         gfp_t mask = GFP_KERNEL;
 126 #endif
 127         return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
 128 }
 129
 130 static inline void free_thread_info(struct thread_info *ti)
 131 {
 132         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 133 }
 134 #endif
 135
 136 /* SLAB cache for signal_struct structures (tsk->signal) */
 137 static struct kmem_cache *signal_cachep;
 138
 139 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 140 struct kmem_cache *sighand_cachep;
 141
 142 /* SLAB cache for files_struct structures (tsk->files) */
 143 struct kmem_cache *files_cachep;
 144
 145 /* SLAB cache for fs_struct structures (tsk->fs) */
 146 struct kmem_cache *fs_cachep;
 147
 148 /* SLAB cache for vm_area_struct structures */
 149 struct kmem_cache *vm_area_cachep;
 150
 151 /* SLAB cache for mm_struct structures (tsk->mm) */
 152 static struct kmem_cache *mm_cachep;
 153
 154 static void account_kernel_stack(struct thread_info *ti, int account)
 155 {
 156         struct zone *zone = page_zone(virt_to_page(ti));
 157
 158         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 159 }
 160
 161 void free_task(struct task_struct *tsk)
 162 {
 163         prop_local_destroy_single(&tsk->dirties);
 164         account_kernel_stack(tsk->stack, -1);
 165         free_thread_info(tsk->stack);
 166         rt_mutex_debug_task_free(tsk);
 167         ftrace_graph_exit_task(tsk);
 168         free_task_struct(tsk);
 169 }
 170 EXPORT_SYMBOL(free_task);
 171
 172 static inline void free_signal_struct(struct signal_struct *sig)
 173 {
 174         taskstats_tgid_free(sig);
 175         sched_autogroup_exit(sig);
 176         kmem_cache_free(signal_cachep, sig);
 177 }
 178
 179 static inline void put_signal_struct(struct signal_struct *sig)
 180 {
 181         if (atomic_dec_and_test(&sig->sigcnt))
 182                 free_signal_struct(sig);
 183 }
 184
 185 void __put_task_struct(struct task_struct *tsk)
 186 {
 187         WARN_ON(!tsk->exit_state);
 188         WARN_ON(atomic_read(&tsk->usage));
 189         WARN_ON(tsk == current);
 190
 191         exit_creds(tsk);
 192         delayacct_tsk_free(tsk);
 193         put_signal_struct(tsk->signal);
 194
 195         if (!profile_handoff_task(tsk))
 196                 free_task(tsk);
 197 }
 198 EXPORT_SYMBOL_GPL(__put_task_struct);
 199
 200 /*
 201  * macro override instead of weak attribute alias, to workaround
 202  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
 203  */
 204 #ifndef arch_task_cache_init
 205 #define arch_task_cache_init()
 206 #endif
 207
 208 void __init fork_init(unsigned long mempages)
 209 {
 210 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 211 #ifndef ARCH_MIN_TASKALIGN
 212 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 213 #endif
 214         /* create a slab on which task_structs can be allocated */
 215         task_struct_cachep =
 216                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 217                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 218 #endif
 219
 220         /* do the arch specific task caches init */
 221         arch_task_cache_init();
 222
 223         /*
 224          * The default maximum number of threads is set to a safe
 225          * value: the thread structures can take up at most half
 226          * of memory.
 227          */
 228         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 229
 230         /*
 231          * we need to allow at least 20 threads to boot a system
 232          */
 233         if(max_threads < 20)
 234                 max_threads = 20;
 235
 236         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 237         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 238         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 239                 init_task.signal->rlim[RLIMIT_NPROC];
 240 }
 241
 242 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 243                                                struct task_struct *src)
 244 {
 245         *dst = *src;
 246         return 0;
 247 }
 248
 249 static struct task_struct *dup_task_struct(struct task_struct *orig)
 250 {
 251         struct task_struct *tsk;
 252         struct thread_info *ti;
 253         unsigned long *stackend;
 254         int node = numa_node_id();
 255         int err;
 256
 257         prepare_to_copy(orig);
 258
 259         tsk = alloc_task_struct_node(node);
 260         if (!tsk)
 261                 return NULL;
 262
 263         ti = alloc_thread_info(tsk);
 264         if (!ti) {
 265                 free_task_struct(tsk);
 266                 return NULL;
 267         }
 268
 269         err = arch_dup_task_struct(tsk, orig);
 270         if (err)
 271                 goto out;
 272
 273         tsk->stack = ti;
 274
 275         err = prop_local_init_single(&tsk->dirties);
 276         if (err)
 277                 goto out;
 278
 279         setup_thread_stack(tsk, orig);
 280         clear_user_return_notifier(tsk);
 281         clear_tsk_need_resched(tsk);
 282         stackend = end_of_stack(tsk);
 283         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 284
 285 #ifdef CONFIG_CC_STACKPROTECTOR
 286         tsk->stack_canary = get_random_int();
 287 #endif
 288
 289         /* One for us, one for whoever does the "release_task()" (usually parent) */
 290         atomic_set(&tsk->usage,2);
 291         atomic_set(&tsk->fs_excl, 0);
 292 #ifdef CONFIG_BLK_DEV_IO_TRACE
 293         tsk->btrace_seq = 0;
 294 #endif
 295         tsk->splice_pipe = NULL;
 296
 297         account_kernel_stack(ti, 1);
 298
 299         return tsk;
 300
 301 out:
 302         free_thread_info(ti);
 303         free_task_struct(tsk);
 304         return NULL;
 305 }
 306
 307 #ifdef CONFIG_MMU
 308 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 309 {
 310         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 311         struct rb_node **rb_link, *rb_parent;
 312         int retval;
 313         unsigned long charge;
 314         struct mempolicy *pol;
 315
 316         down_write(&oldmm->mmap_sem);
 317         flush_cache_dup_mm(oldmm);
 318         /*
 319          * Not linked in yet - no deadlock potential:
 320          */
 321         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 322
 323         mm->locked_vm = 0;
 324         mm->mmap = NULL;
 325         mm->mmap_cache = NULL;
 326         mm->free_area_cache = oldmm->mmap_base;
 327         mm->cached_hole_size = ~0UL;
 328         mm->map_count = 0;
 329         cpumask_clear(mm_cpumask(mm));
 330         mm->mm_rb = RB_ROOT;
 331         rb_link = &mm->mm_rb.rb_node;
 332         rb_parent = NULL;
 333         pprev = &mm->mmap;
 334         retval = ksm_fork(mm, oldmm);
 335         if (retval)
 336                 goto out;
 337         retval = khugepaged_fork(mm, oldmm);
 338         if (retval)
 339                 goto out;
 340
 341         prev = NULL;
 342         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 343                 struct file *file;
 344
 345                 if (mpnt->vm_flags & VM_DONTCOPY) {
 346                         long pages = vma_pages(mpnt);
 347                         mm->total_vm -= pages;
 348                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 349                                                                 -pages);
 350                         continue;
 351                 }
 352                 charge = 0;
 353                 if (mpnt->vm_flags & VM_ACCOUNT) {
 354                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 355                         if (security_vm_enough_memory(len))
 356                                 goto fail_nomem;
 357                         charge = len;
 358                 }
 359                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 360                 if (!tmp)
 361                         goto fail_nomem;
 362                 *tmp = *mpnt;
 363                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 364                 pol = mpol_dup(vma_policy(mpnt));
 365                 retval = PTR_ERR(pol);
 366                 if (IS_ERR(pol))
 367                         goto fail_nomem_policy;
 368                 vma_set_policy(tmp, pol);
 369                 tmp->vm_mm = mm;
 370                 if (anon_vma_fork(tmp, mpnt))
 371                         goto fail_nomem_anon_vma_fork;
 372                 tmp->vm_flags &= ~VM_LOCKED;
 373                 tmp->vm_next = tmp->vm_prev = NULL;
 374                 file = tmp->vm_file;
 375                 if (file) {
 376                         struct inode *inode = file->f_path.dentry->d_inode;
 377                         struct address_space *mapping = file->f_mapping;
 378
 379                         get_file(file);
 380                         if (tmp->vm_flags & VM_DENYWRITE)
 381                                 atomic_dec(&inode->i_writecount);
 382                         spin_lock(&mapping->i_mmap_lock);
 383                         if (tmp->vm_flags & VM_SHARED)
 384                                 mapping->i_mmap_writable++;
 385                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
 386                         flush_dcache_mmap_lock(mapping);
 387                         /* insert tmp into the share list, just after mpnt */
 388                         vma_prio_tree_add(tmp, mpnt);
 389                         flush_dcache_mmap_unlock(mapping);
 390                         spin_unlock(&mapping->i_mmap_lock);
 391                 }
 392
 393                 /*
 394                  * Clear hugetlb-related page reserves for children. This only
 395                  * affects MAP_PRIVATE mappings. Faults generated by the child
 396                  * are not guaranteed to succeed, even if read-only
 397                  */
 398                 if (is_vm_hugetlb_page(tmp))
 399                         reset_vma_resv_huge_pages(tmp);
 400
 401                 /*
 402                  * Link in the new vma and copy the page table entries.
 403                  */
 404                 *pprev = tmp;
 405                 pprev = &tmp->vm_next;
 406                 tmp->vm_prev = prev;
 407                 prev = tmp;
 408
 409                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 410                 rb_link = &tmp->vm_rb.rb_right;
 411                 rb_parent = &tmp->vm_rb;
 412
 413                 mm->map_count++;
 414                 retval = copy_page_range(mm, oldmm, mpnt);
 415
 416                 if (tmp->vm_ops && tmp->vm_ops->open)
 417                         tmp->vm_ops->open(tmp);
 418
 419                 if (retval)
 420                         goto out;
 421         }
 422         /* a new mm has just been created */
 423         arch_dup_mmap(oldmm, mm);
 424         retval = 0;
 425 out:
 426         up_write(&mm->mmap_sem);
 427         flush_tlb_mm(oldmm);
 428         up_write(&oldmm->mmap_sem);
 429         return retval;
 430 fail_nomem_anon_vma_fork:
 431         mpol_put(pol);
 432 fail_nomem_policy:
 433         kmem_cache_free(vm_area_cachep, tmp);
 434 fail_nomem:
 435         retval = -ENOMEM;
 436         vm_unacct_memory(charge);
 437         goto out;
 438 }
 439
 440 static inline int mm_alloc_pgd(struct mm_struct * mm)
 441 {
 442         mm->pgd = pgd_alloc(mm);
 443         if (unlikely(!mm->pgd))
 444                 return -ENOMEM;
 445         return 0;
 446 }
 447
 448 static inline void mm_free_pgd(struct mm_struct * mm)
 449 {
 450         pgd_free(mm, mm->pgd);
 451 }
 452 #else
 453 #define dup_mmap(mm, oldmm)     (0)
 454 #define mm_alloc_pgd(mm)        (0)
 455 #define mm_free_pgd(mm)
 456 #endif /* CONFIG_MMU */
 457
 458 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 459
 460 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 461 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 462
 463 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 464
 465 static int __init coredump_filter_setup(char *s)
 466 {
 467         default_dump_filter =
 468                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 469                 MMF_DUMP_FILTER_MASK;
 470         return 1;
 471 }
 472
 473 __setup("coredump_filter=", coredump_filter_setup);
 474
 475 #include <linux/init_task.h>
 476
 477 static void mm_init_aio(struct mm_struct *mm)
 478 {
 479 #ifdef CONFIG_AIO
 480         spin_lock_init(&mm->ioctx_lock);
 481         INIT_HLIST_HEAD(&mm->ioctx_list);
 482 #endif
 483 }
 484
 485 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
 486 {
 487         atomic_set(&mm->mm_users, 1);
 488         atomic_set(&mm->mm_count, 1);
 489         init_rwsem(&mm->mmap_sem);
 490         INIT_LIST_HEAD(&mm->mmlist);
 491         mm->flags = (current->mm) ?
 492                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
 493         mm->core_state = NULL;
 494         mm->nr_ptes = 0;
 495         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 496         spin_lock_init(&mm->page_table_lock);
 497         mm->free_area_cache = TASK_UNMAPPED_BASE;
 498         mm->cached_hole_size = ~0UL;
 499         mm_init_aio(mm);
 500         mm_init_owner(mm, p);
 501         atomic_set(&mm->oom_disable_count, 0);
 502
 503         if (likely(!mm_alloc_pgd(mm))) {
 504                 mm->def_flags = 0;
 505                 mmu_notifier_mm_init(mm);
 506                 return mm;
 507         }
 508
 509         free_mm(mm);
 510         return NULL;
 511 }
 512
 513 /*
 514  * Allocate and initialize an mm_struct.
 515  */
 516 struct mm_struct * mm_alloc(void)
 517 {
 518         struct mm_struct * mm;
 519
 520         mm = allocate_mm();
 521         if (mm) {
 522                 memset(mm, 0, sizeof(*mm));
 523                 mm = mm_init(mm, current);
 524         }
 525         return mm;
 526 }
 527
 528 /*
 529  * Called when the last reference to the mm
 530  * is dropped: either by a lazy thread or by
 531  * mmput. Free the page directory and the mm.
 532  */
 533 void __mmdrop(struct mm_struct *mm)
 534 {
 535         BUG_ON(mm == &init_mm);
 536         mm_free_pgd(mm);
 537         destroy_context(mm);
 538         mmu_notifier_mm_destroy(mm);
 539 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 540         VM_BUG_ON(mm->pmd_huge_pte);
 541 #endif
 542         free_mm(mm);
 543 }
 544 EXPORT_SYMBOL_GPL(__mmdrop);
 545
 546 /*
 547  * Decrement the use count and release all resources for an mm.
 548  */
 549 void mmput(struct mm_struct *mm)
 550 {
 551         might_sleep();
 552
 553         if (atomic_dec_and_test(&mm->mm_users)) {
 554                 exit_aio(mm);
 555                 ksm_exit(mm);
 556                 khugepaged_exit(mm); /* must run before exit_mmap */
 557                 exit_mmap(mm);
 558                 set_mm_exe_file(mm, NULL);
 559                 if (!list_empty(&mm->mmlist)) {
 560                         spin_lock(&mmlist_lock);
 561                         list_del(&mm->mmlist);
 562                         spin_unlock(&mmlist_lock);
 563                 }
 564                 put_swap_token(mm);
 565                 if (mm->binfmt)
 566                         module_put(mm->binfmt->module);
 567                 mmdrop(mm);
 568         }
 569 }
 570 EXPORT_SYMBOL_GPL(mmput);
 571
 572 /**
 573  * get_task_mm - acquire a reference to the task's mm
 574  *
 575  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 576  * this kernel workthread has transiently adopted a user mm with use_mm,
 577  * to do its AIO) is not set and if so returns a reference to it, after
 578  * bumping up the use count.  User must release the mm via mmput()
 579  * after use.  Typically used by /proc and ptrace.
 580  */
 581 struct mm_struct *get_task_mm(struct task_struct *task)
 582 {
 583         struct mm_struct *mm;
 584
 585         task_lock(task);
 586         mm = task->mm;
 587         if (mm) {
 588                 if (task->flags & PF_KTHREAD)
 589                         mm = NULL;
 590                 else
 591                         atomic_inc(&mm->mm_users);
 592         }
 593         task_unlock(task);
 594         return mm;
 595 }
 596 EXPORT_SYMBOL_GPL(get_task_mm);
 597
 598 /* Please note the differences between mmput and mm_release.
 599  * mmput is called whenever we stop holding onto a mm_struct,
 600  * error success whatever.
 601  *
 602  * mm_release is called after a mm_struct has been removed
 603  * from the current process.
 604  *
 605  * This difference is important for error handling, when we
 606  * only half set up a mm_struct for a new process and need to restore
 607  * the old one.  Because we mmput the new mm_struct before
 608  * restoring the old one. . .
 609  * Eric Biederman 10 January 1998
 610  */
 611 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 612 {
 613         struct completion *vfork_done = tsk->vfork_done;
 614
 615         /* Get rid of any futexes when releasing the mm */
 616 #ifdef CONFIG_FUTEX
 617         if (unlikely(tsk->robust_list)) {
 618                 exit_robust_list(tsk);
 619                 tsk->robust_list = NULL;
 620         }
 621 #ifdef CONFIG_COMPAT
 622         if (unlikely(tsk->compat_robust_list)) {
 623                 compat_exit_robust_list(tsk);
 624                 tsk->compat_robust_list = NULL;
 625         }
 626 #endif
 627         if (unlikely(!list_empty(&tsk->pi_state_list)))
 628                 exit_pi_state_list(tsk);
 629 #endif
 630
 631         /* Get rid of any cached register state */
 632         deactivate_mm(tsk, mm);
 633
 634         /* notify parent sleeping on vfork() */
 635         if (vfork_done) {
 636                 tsk->vfork_done = NULL;
 637                 complete(vfork_done);
 638         }
 639
 640         /*
 641          * If we're exiting normally, clear a user-space tid field if
 642          * requested.  We leave this alone when dying by signal, to leave
 643          * the value intact in a core dump, and to save the unnecessary
 644          * trouble otherwise.  Userland only wants this done for a sys_exit.
 645          */
 646         if (tsk->clear_child_tid) {
 647                 if (!(tsk->flags & PF_SIGNALED) &&
 648                     atomic_read(&mm->mm_users) > 1) {
 649                         /*
 650                          * We don't check the error code - if userspace has
 651                          * not set up a proper pointer then tough luck.
 652                          */
 653                         put_user(0, tsk->clear_child_tid);
 654                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 655                                         1, NULL, NULL, 0);
 656                 }
 657                 tsk->clear_child_tid = NULL;
 658         }
 659 }
 660
 661 /*
 662  * Allocate a new mm structure and copy contents from the
 663  * mm structure of the passed in task structure.
 664  */
 665 struct mm_struct *dup_mm(struct task_struct *tsk)
 666 {
 667         struct mm_struct *mm, *oldmm = current->mm;
 668         int err;
 669
 670         if (!oldmm)
 671                 return NULL;
 672
 673         mm = allocate_mm();
 674         if (!mm)
 675                 goto fail_nomem;
 676
 677         memcpy(mm, oldmm, sizeof(*mm));
 678
 679         /* Initializing for Swap token stuff */
 680         mm->token_priority = 0;
 681         mm->last_interval = 0;
 682
 683 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 684         mm->pmd_huge_pte = NULL;
 685 #endif
 686
 687         if (!mm_init(mm, tsk))
 688                 goto fail_nomem;
 689
 690         if (init_new_context(tsk, mm))
 691                 goto fail_nocontext;
 692
 693         dup_mm_exe_file(oldmm, mm);
 694
 695         err = dup_mmap(mm, oldmm);
 696         if (err)
 697                 goto free_pt;
 698
 699         mm->hiwater_rss = get_mm_rss(mm);
 700         mm->hiwater_vm = mm->total_vm;
 701
 702         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 703                 goto free_pt;
 704
 705         return mm;
 706
 707 free_pt:
 708         /* don't put binfmt in mmput, we haven't got module yet */
 709         mm->binfmt = NULL;
 710         mmput(mm);
 711
 712 fail_nomem:
 713         return NULL;
 714
 715 fail_nocontext:
 716         /*
 717          * If init_new_context() failed, we cannot use mmput() to free the mm
 718          * because it calls destroy_context()
 719          */
 720         mm_free_pgd(mm);
 721         free_mm(mm);
 722         return NULL;
 723 }
 724
 725 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 726 {
 727         struct mm_struct * mm, *oldmm;
 728         int retval;
 729
 730         tsk->min_flt = tsk->maj_flt = 0;
 731         tsk->nvcsw = tsk->nivcsw = 0;
 732 #ifdef CONFIG_DETECT_HUNG_TASK
 733         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 734 #endif
 735
 736         tsk->mm = NULL;
 737         tsk->active_mm = NULL;
 738
 739         /*
 740          * Are we cloning a kernel thread?
 741          *
 742          * We need to steal a active VM for that..
 743          */
 744         oldmm = current->mm;
 745         if (!oldmm)
 746                 return 0;
 747
 748         if (clone_flags & CLONE_VM) {
 749                 atomic_inc(&oldmm->mm_users);
 750                 mm = oldmm;
 751                 goto good_mm;
 752         }
 753
 754         retval = -ENOMEM;
 755         mm = dup_mm(tsk);
 756         if (!mm)
 757                 goto fail_nomem;
 758
 759 good_mm:
 760         /* Initializing for Swap token stuff */
 761         mm->token_priority = 0;
 762         mm->last_interval = 0;
 763         if (tsk->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
 764                 atomic_inc(&mm->oom_disable_count);
 765
 766         tsk->mm = mm;
 767         tsk->active_mm = mm;
 768         return 0;
 769
 770 fail_nomem:
 771         return retval;
 772 }
 773
 774 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 775 {
 776         struct fs_struct *fs = current->fs;
 777         if (clone_flags & CLONE_FS) {
 778                 /* tsk->fs is already what we want */
 779                 spin_lock(&fs->lock);
 780                 if (fs->in_exec) {
 781                         spin_unlock(&fs->lock);
 782                         return -EAGAIN;
 783                 }
 784                 fs->users++;
 785                 spin_unlock(&fs->lock);
 786                 return 0;
 787         }
 788         tsk->fs = copy_fs_struct(fs);
 789         if (!tsk->fs)
 790                 return -ENOMEM;
 791         return 0;
 792 }
 793
 794 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 795 {
 796         struct files_struct *oldf, *newf;
 797         int error = 0;
 798
 799         /*
 800          * A background process may not have any files ...
 801          */
 802         oldf = current->files;
 803         if (!oldf)
 804                 goto out;
 805
 806         if (clone_flags & CLONE_FILES) {
 807                 atomic_inc(&oldf->count);
 808                 goto out;
 809         }
 810
 811         newf = dup_fd(oldf, &error);
 812         if (!newf)
 813                 goto out;
 814
 815         tsk->files = newf;
 816         error = 0;
 817 out:
 818         return error;
 819 }
 820
 821 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 822 {
 823 #ifdef CONFIG_BLOCK
 824         struct io_context *ioc = current->io_context;
 825
 826         if (!ioc)
 827                 return 0;
 828         /*
 829          * Share io context with parent, if CLONE_IO is set
 830          */
 831         if (clone_flags & CLONE_IO) {
 832                 tsk->io_context = ioc_task_link(ioc);
 833                 if (unlikely(!tsk->io_context))
 834                         return -ENOMEM;
 835         } else if (ioprio_valid(ioc->ioprio)) {
 836                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
 837                 if (unlikely(!tsk->io_context))
 838                         return -ENOMEM;
 839
 840                 tsk->io_context->ioprio = ioc->ioprio;
 841         }
 842 #endif
 843         return 0;
 844 }
 845
 846 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 847 {
 848         struct sighand_struct *sig;
 849
 850         if (clone_flags & CLONE_SIGHAND) {
 851                 atomic_inc(&current->sighand->count);
 852                 return 0;
 853         }
 854         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 855         rcu_assign_pointer(tsk->sighand, sig);
 856         if (!sig)
 857                 return -ENOMEM;
 858         atomic_set(&sig->count, 1);
 859         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 860         return 0;
 861 }
 862
 863 void __cleanup_sighand(struct sighand_struct *sighand)
 864 {
 865         if (atomic_dec_and_test(&sighand->count))
 866                 kmem_cache_free(sighand_cachep, sighand);
 867 }
 868
 869
 870 /*
 871  * Initialize POSIX timer handling for a thread group.
 872  */
 873 static void posix_cpu_timers_init_group(struct signal_struct *sig)
 874 {
 875         unsigned long cpu_limit;
 876
 877         /* Thread group counters. */
 878         thread_group_cputime_init(sig);
 879
 880         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
 881         if (cpu_limit != RLIM_INFINITY) {
 882                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
 883                 sig->cputimer.running = 1;
 884         }
 885
 886         /* The timer lists. */
 887         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 888         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 889         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 890 }
 891
 892 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
 893 {
 894         struct signal_struct *sig;
 895
 896         if (clone_flags & CLONE_THREAD)
 897                 return 0;
 898
 899         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
 900         tsk->signal = sig;
 901         if (!sig)
 902                 return -ENOMEM;
 903
 904         sig->nr_threads = 1;
 905         atomic_set(&sig->live, 1);
 906         atomic_set(&sig->sigcnt, 1);
 907         init_waitqueue_head(&sig->wait_chldexit);
 908         if (clone_flags & CLONE_NEWPID)
 909                 sig->flags |= SIGNAL_UNKILLABLE;
 910         sig->curr_target = tsk;
 911         init_sigpending(&sig->shared_pending);
 912         INIT_LIST_HEAD(&sig->posix_timers);
 913
 914         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 915         sig->real_timer.function = it_real_fn;
 916
 917         task_lock(current->group_leader);
 918         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 919         task_unlock(current->group_leader);
 920
 921         posix_cpu_timers_init_group(sig);
 922
 923         tty_audit_fork(sig);
 924         sched_autogroup_fork(sig);
 925
 926         sig->oom_adj = current->signal->oom_adj;
 927         sig->oom_score_adj = current->signal->oom_score_adj;
 928         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
 929
 930         mutex_init(&sig->cred_guard_mutex);
 931
 932         return 0;
 933 }
 934
 935 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
 936 {
 937         unsigned long new_flags = p->flags;
 938
 939         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
 940         new_flags |= PF_FORKNOEXEC;
 941         new_flags |= PF_STARTING;
 942         p->flags = new_flags;
 943         clear_freeze_flag(p);
 944 }
 945
 946 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
 947 {
 948         current->clear_child_tid = tidptr;
 949
 950         return task_pid_vnr(current);
 951 }
 952
 953 static void rt_mutex_init_task(struct task_struct *p)
 954 {
 955         raw_spin_lock_init(&p->pi_lock);
 956 #ifdef CONFIG_RT_MUTEXES
 957         plist_head_init_raw(&p->pi_waiters, &p->pi_lock);
 958         p->pi_blocked_on = NULL;
 959 #endif
 960 }
 961
 962 #ifdef CONFIG_MM_OWNER
 963 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 964 {
 965         mm->owner = p;
 966 }
 967 #endif /* CONFIG_MM_OWNER */
 968
 969 /*
 970  * Initialize POSIX timer handling for a single task.
 971  */
 972 static void posix_cpu_timers_init(struct task_struct *tsk)
 973 {
 974         tsk->cputime_expires.prof_exp = cputime_zero;
 975         tsk->cputime_expires.virt_exp = cputime_zero;
 976         tsk->cputime_expires.sched_exp = 0;
 977         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
 978         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
 979         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
 980 }
 981
 982 /*
 983  * This creates a new process as a copy of the old one,
 984  * but does not actually start it yet.
 985  *
 986  * It copies the registers, and all the appropriate
 987  * parts of the process environment (as per the clone
 988  * flags). The actual kick-off is left to the caller.
 989  */
 990 static struct task_struct *copy_process(unsigned long clone_flags,
 991                                         unsigned long stack_start,
 992                                         struct pt_regs *regs,
 993                                         unsigned long stack_size,
 994                                         int __user *child_tidptr,
 995                                         struct pid *pid,
 996                                         int trace)
 997 {
 998         int retval;
 999         struct task_struct *p;
1000         int cgroup_callbacks_done = 0;
1001
1002         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1003                 return ERR_PTR(-EINVAL);
1004
1005         /*
1006          * Thread groups must share signals as well, and detached threads
1007          * can only be started up within the thread group.
1008          */
1009         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1010                 return ERR_PTR(-EINVAL);
1011
1012         /*
1013          * Shared signal handlers imply shared VM. By way of the above,
1014          * thread groups also imply shared VM. Blocking this case allows
1015          * for various simplifications in other code.
1016          */
1017         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1018                 return ERR_PTR(-EINVAL);
1019
1020         /*
1021          * Siblings of global init remain as zombies on exit since they are
1022          * not reaped by their parent (swapper). To solve this and to avoid
1023          * multi-rooted process trees, prevent global and container-inits
1024          * from creating siblings.
1025          */
1026         if ((clone_flags & CLONE_PARENT) &&
1027                                 current->signal->flags & SIGNAL_UNKILLABLE)
1028                 return ERR_PTR(-EINVAL);
1029
1030         retval = security_task_create(clone_flags);
1031         if (retval)
1032                 goto fork_out;
1033
1034         retval = -ENOMEM;
1035         p = dup_task_struct(current);
1036         if (!p)
1037                 goto fork_out;
1038
1039         ftrace_graph_init_task(p);
1040
1041         rt_mutex_init_task(p);
1042
1043 #ifdef CONFIG_PROVE_LOCKING
1044         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1045         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1046 #endif
1047         retval = -EAGAIN;
1048         if (atomic_read(&p->real_cred->user->processes) >=
1049                         task_rlimit(p, RLIMIT_NPROC)) {
1050                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1051                     p->real_cred->user != INIT_USER)
1052                         goto bad_fork_free;
1053         }
1054
1055         retval = copy_creds(p, clone_flags);
1056         if (retval < 0)
1057                 goto bad_fork_free;
1058
1059         /*
1060          * If multiple threads are within copy_process(), then this check
1061          * triggers too late. This doesn't hurt, the check is only there
1062          * to stop root fork bombs.
1063          */
1064         retval = -EAGAIN;
1065         if (nr_threads >= max_threads)
1066                 goto bad_fork_cleanup_count;
1067
1068         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1069                 goto bad_fork_cleanup_count;
1070
1071         p->did_exec = 0;
1072         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1073         copy_flags(clone_flags, p);
1074         INIT_LIST_HEAD(&p->children);
1075         INIT_LIST_HEAD(&p->sibling);
1076         rcu_copy_process(p);
1077         p->vfork_done = NULL;
1078         spin_lock_init(&p->alloc_lock);
1079
1080         init_sigpending(&p->pending);
1081
1082         p->utime = cputime_zero;
1083         p->stime = cputime_zero;
1084         p->gtime = cputime_zero;
1085         p->utimescaled = cputime_zero;
1086         p->stimescaled = cputime_zero;
1087 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1088         p->prev_utime = cputime_zero;
1089         p->prev_stime = cputime_zero;
1090 #endif
1091 #if defined(SPLIT_RSS_COUNTING)
1092         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1093 #endif
1094
1095         p->default_timer_slack_ns = current->timer_slack_ns;
1096
1097         task_io_accounting_init(&p->ioac);
1098         acct_clear_integrals(p);
1099
1100         posix_cpu_timers_init(p);
1101
1102         p->lock_depth = -1;             /* -1 = no lock */
1103         do_posix_clock_monotonic_gettime(&p->start_time);
1104         p->real_start_time = p->start_time;
1105         monotonic_to_bootbased(&p->real_start_time);
1106         p->io_context = NULL;
1107         p->audit_context = NULL;
1108         cgroup_fork(p);
1109 #ifdef CONFIG_NUMA
1110         p->mempolicy = mpol_dup(p->mempolicy);
1111         if (IS_ERR(p->mempolicy)) {
1112                 retval = PTR_ERR(p->mempolicy);
1113                 p->mempolicy = NULL;
1114                 goto bad_fork_cleanup_cgroup;
1115         }
1116         mpol_fix_fork_child_flag(p);
1117 #endif
1118 #ifdef CONFIG_TRACE_IRQFLAGS
1119         p->irq_events = 0;
1120 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1121         p->hardirqs_enabled = 1;
1122 #else
1123         p->hardirqs_enabled = 0;
1124 #endif
1125         p->hardirq_enable_ip = 0;
1126         p->hardirq_enable_event = 0;
1127         p->hardirq_disable_ip = _THIS_IP_;
1128         p->hardirq_disable_event = 0;
1129         p->softirqs_enabled = 1;
1130         p->softirq_enable_ip = _THIS_IP_;
1131         p->softirq_enable_event = 0;
1132         p->softirq_disable_ip = 0;
1133         p->softirq_disable_event = 0;
1134         p->hardirq_context = 0;
1135         p->softirq_context = 0;
1136 #endif
1137 #ifdef CONFIG_LOCKDEP
1138         p->lockdep_depth = 0; /* no locks held yet */
1139         p->curr_chain_key = 0;
1140         p->lockdep_recursion = 0;
1141 #endif
1142
1143 #ifdef CONFIG_DEBUG_MUTEXES
1144         p->blocked_on = NULL; /* not blocked yet */
1145 #endif
1146 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1147         p->memcg_batch.do_batch = 0;
1148         p->memcg_batch.memcg = NULL;
1149 #endif
1150
1151         /* Perform scheduler related setup. Assign this task to a CPU. */
1152         sched_fork(p, clone_flags);
1153
1154         retval = perf_event_init_task(p);
1155         if (retval)
1156                 goto bad_fork_cleanup_policy;
1157
1158         if ((retval = audit_alloc(p)))
1159                 goto bad_fork_cleanup_policy;
1160         /* copy all the process information */
1161         if ((retval = copy_semundo(clone_flags, p)))
1162                 goto bad_fork_cleanup_audit;
1163         if ((retval = copy_files(clone_flags, p)))
1164                 goto bad_fork_cleanup_semundo;
1165         if ((retval = copy_fs(clone_flags, p)))
1166                 goto bad_fork_cleanup_files;
1167         if ((retval = copy_sighand(clone_flags, p)))
1168                 goto bad_fork_cleanup_fs;
1169         if ((retval = copy_signal(clone_flags, p)))
1170                 goto bad_fork_cleanup_sighand;
1171         if ((retval = copy_mm(clone_flags, p)))
1172                 goto bad_fork_cleanup_signal;
1173         if ((retval = copy_namespaces(clone_flags, p)))
1174                 goto bad_fork_cleanup_mm;
1175         if ((retval = copy_io(clone_flags, p)))
1176                 goto bad_fork_cleanup_namespaces;
1177         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1178         if (retval)
1179                 goto bad_fork_cleanup_io;
1180
1181         if (pid != &init_struct_pid) {
1182                 retval = -ENOMEM;
1183                 pid = alloc_pid(p->nsproxy->pid_ns);
1184                 if (!pid)
1185                         goto bad_fork_cleanup_io;
1186
1187                 if (clone_flags & CLONE_NEWPID) {
1188                         retval = pid_ns_prepare_proc(p->nsproxy->pid_ns);
1189                         if (retval < 0)
1190                                 goto bad_fork_free_pid;
1191                 }
1192         }
1193
1194         p->pid = pid_nr(pid);
1195         p->tgid = p->pid;
1196         if (clone_flags & CLONE_THREAD)
1197                 p->tgid = current->tgid;
1198
1199         if (current->nsproxy != p->nsproxy) {
1200                 retval = ns_cgroup_clone(p, pid);
1201                 if (retval)
1202                         goto bad_fork_free_pid;
1203         }
1204
1205         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1206         /*
1207          * Clear TID on mm_release()?
1208          */
1209         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1210 #ifdef CONFIG_FUTEX
1211         p->robust_list = NULL;
1212 #ifdef CONFIG_COMPAT
1213         p->compat_robust_list = NULL;
1214 #endif
1215         INIT_LIST_HEAD(&p->pi_state_list);
1216         p->pi_state_cache = NULL;
1217 #endif
1218         /*
1219          * sigaltstack should be cleared when sharing the same VM
1220          */
1221         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1222                 p->sas_ss_sp = p->sas_ss_size = 0;
1223
1224         /*
1225          * Syscall tracing and stepping should be turned off in the
1226          * child regardless of CLONE_PTRACE.
1227          */
1228         user_disable_single_step(p);
1229         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1230 #ifdef TIF_SYSCALL_EMU
1231         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1232 #endif
1233         clear_all_latency_tracing(p);
1234
1235         /* ok, now we should be set up.. */
1236         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1237         p->pdeath_signal = 0;
1238         p->exit_state = 0;
1239
1240         /*
1241          * Ok, make it visible to the rest of the system.
1242          * We dont wake it up yet.
1243          */
1244         p->group_leader = p;
1245         INIT_LIST_HEAD(&p->thread_group);
1246
1247         /* Now that the task is set up, run cgroup callbacks if
1248          * necessary. We need to run them before the task is visible
1249          * on the tasklist. */
1250         cgroup_fork_callbacks(p);
1251         cgroup_callbacks_done = 1;
1252
1253         /* Need tasklist lock for parent etc handling! */
1254         write_lock_irq(&tasklist_lock);
1255
1256         /* CLONE_PARENT re-uses the old parent */
1257         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1258                 p->real_parent = current->real_parent;
1259                 p->parent_exec_id = current->parent_exec_id;
1260         } else {
1261                 p->real_parent = current;
1262                 p->parent_exec_id = current->self_exec_id;
1263         }
1264
1265         spin_lock(&current->sighand->siglock);
1266
1267         /*
1268          * Process group and session signals need to be delivered to just the
1269          * parent before the fork or both the parent and the child after the
1270          * fork. Restart if a signal comes in before we add the new process to
1271          * it's process group.
1272          * A fatal signal pending means that current will exit, so the new
1273          * thread can't slip out of an OOM kill (or normal SIGKILL).
1274          */
1275         recalc_sigpending();
1276         if (signal_pending(current)) {
1277                 spin_unlock(&current->sighand->siglock);
1278                 write_unlock_irq(&tasklist_lock);
1279                 retval = -ERESTARTNOINTR;
1280                 goto bad_fork_free_pid;
1281         }
1282
1283         if (clone_flags & CLONE_THREAD) {
1284                 current->signal->nr_threads++;
1285                 atomic_inc(&current->signal->live);
1286                 atomic_inc(&current->signal->sigcnt);
1287                 p->group_leader = current->group_leader;
1288                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1289         }
1290
1291         if (likely(p->pid)) {
1292                 tracehook_finish_clone(p, clone_flags, trace);
1293
1294                 if (thread_group_leader(p)) {
1295                         if (clone_flags & CLONE_NEWPID)
1296                                 p->nsproxy->pid_ns->child_reaper = p;
1297
1298                         p->signal->leader_pid = pid;
1299                         p->signal->tty = tty_kref_get(current->signal->tty);
1300                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1301                         attach_pid(p, PIDTYPE_SID, task_session(current));
1302                         list_add_tail(&p->sibling, &p->real_parent->children);
1303                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1304                         __this_cpu_inc(process_counts);
1305                 }
1306                 attach_pid(p, PIDTYPE_PID, pid);
1307                 nr_threads++;
1308         }
1309
1310         total_forks++;
1311         spin_unlock(&current->sighand->siglock);
1312         write_unlock_irq(&tasklist_lock);
1313         proc_fork_connector(p);
1314         cgroup_post_fork(p);
1315         perf_event_fork(p);
1316         return p;
1317
1318 bad_fork_free_pid:
1319         if (pid != &init_struct_pid)
1320                 free_pid(pid);
1321 bad_fork_cleanup_io:
1322         if (p->io_context)
1323                 exit_io_context(p);
1324 bad_fork_cleanup_namespaces:
1325         exit_task_namespaces(p);
1326 bad_fork_cleanup_mm:
1327         if (p->mm) {
1328                 task_lock(p);
1329                 if (p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1330                         atomic_dec(&p->mm->oom_disable_count);
1331                 task_unlock(p);
1332                 mmput(p->mm);
1333         }
1334 bad_fork_cleanup_signal:
1335         if (!(clone_flags & CLONE_THREAD))
1336                 free_signal_struct(p->signal);
1337 bad_fork_cleanup_sighand:
1338         __cleanup_sighand(p->sighand);
1339 bad_fork_cleanup_fs:
1340         exit_fs(p); /* blocking */
1341 bad_fork_cleanup_files:
1342         exit_files(p); /* blocking */
1343 bad_fork_cleanup_semundo:
1344         exit_sem(p);
1345 bad_fork_cleanup_audit:
1346         audit_free(p);
1347 bad_fork_cleanup_policy:
1348         perf_event_free_task(p);
1349 #ifdef CONFIG_NUMA
1350         mpol_put(p->mempolicy);
1351 bad_fork_cleanup_cgroup:
1352 #endif
1353         cgroup_exit(p, cgroup_callbacks_done);
1354         delayacct_tsk_free(p);
1355         module_put(task_thread_info(p)->exec_domain->module);
1356 bad_fork_cleanup_count:
1357         atomic_dec(&p->cred->user->processes);
1358         exit_creds(p);
1359 bad_fork_free:
1360         free_task(p);
1361 fork_out:
1362         return ERR_PTR(retval);
1363 }
1364
1365 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1366 {
1367         memset(regs, 0, sizeof(struct pt_regs));
1368         return regs;
1369 }
1370
1371 static inline void init_idle_pids(struct pid_link *links)
1372 {
1373         enum pid_type type;
1374
1375         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1376                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1377                 links[type].pid = &init_struct_pid;
1378         }
1379 }
1380
1381 struct task_struct * __cpuinit fork_idle(int cpu)
1382 {
1383         struct task_struct *task;
1384         struct pt_regs regs;
1385
1386         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1387                             &init_struct_pid, 0);
1388         if (!IS_ERR(task)) {
1389                 init_idle_pids(task->pids);
1390                 init_idle(task, cpu);
1391         }
1392
1393         return task;
1394 }
1395
1396 /*
1397  *  Ok, this is the main fork-routine.
1398  *
1399  * It copies the process, and if successful kick-starts
1400  * it and waits for it to finish using the VM if required.
1401  */
1402 long do_fork(unsigned long clone_flags,
1403               unsigned long stack_start,
1404               struct pt_regs *regs,
1405               unsigned long stack_size,
1406               int __user *parent_tidptr,
1407               int __user *child_tidptr)
1408 {
1409         struct task_struct *p;
1410         int trace = 0;
1411         long nr;
1412
1413         /*
1414          * Do some preliminary argument and permissions checking before we
1415          * actually start allocating stuff
1416          */
1417         if (clone_flags & CLONE_NEWUSER) {
1418                 if (clone_flags & CLONE_THREAD)
1419                         return -EINVAL;
1420                 /* hopefully this check will go away when userns support is
1421                  * complete
1422                  */
1423                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1424                                 !capable(CAP_SETGID))
1425                         return -EPERM;
1426         }
1427
1428         /*
1429          * When called from kernel_thread, don't do user tracing stuff.
1430          */
1431         if (likely(user_mode(regs)))
1432                 trace = tracehook_prepare_clone(clone_flags);
1433
1434         p = copy_process(clone_flags, stack_start, regs, stack_size,
1435                          child_tidptr, NULL, trace);
1436         /*
1437          * Do this prior waking up the new thread - the thread pointer
1438          * might get invalid after that point, if the thread exits quickly.
1439          */
1440         if (!IS_ERR(p)) {
1441                 struct completion vfork;
1442
1443                 trace_sched_process_fork(current, p);
1444
1445                 nr = task_pid_vnr(p);
1446
1447                 if (clone_flags & CLONE_PARENT_SETTID)
1448                         put_user(nr, parent_tidptr);
1449
1450                 if (clone_flags & CLONE_VFORK) {
1451                         p->vfork_done = &vfork;
1452                         init_completion(&vfork);
1453                 }
1454
1455                 audit_finish_fork(p);
1456                 tracehook_report_clone(regs, clone_flags, nr, p);
1457
1458                 /*
1459                  * We set PF_STARTING at creation in case tracing wants to
1460                  * use this to distinguish a fully live task from one that
1461                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1462                  * clear it and set the child going.
1463                  */
1464                 p->flags &= ~PF_STARTING;
1465
1466                 wake_up_new_task(p, clone_flags);
1467
1468                 tracehook_report_clone_complete(trace, regs,
1469                                                 clone_flags, nr, p);
1470
1471                 if (clone_flags & CLONE_VFORK) {
1472                         freezer_do_not_count();
1473                         wait_for_completion(&vfork);
1474                         freezer_count();
1475                         tracehook_report_vfork_done(p, nr);
1476                 }
1477         } else {
1478                 nr = PTR_ERR(p);
1479         }
1480         return nr;
1481 }
1482
1483 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1484 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1485 #endif
1486
1487 static void sighand_ctor(void *data)
1488 {
1489         struct sighand_struct *sighand = data;
1490
1491         spin_lock_init(&sighand->siglock);
1492         init_waitqueue_head(&sighand->signalfd_wqh);
1493 }
1494
1495 void __init proc_caches_init(void)
1496 {
1497         sighand_cachep = kmem_cache_create("sighand_cache",
1498                         sizeof(struct sighand_struct), 0,
1499                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1500                         SLAB_NOTRACK, sighand_ctor);
1501         signal_cachep = kmem_cache_create("signal_cache",
1502                         sizeof(struct signal_struct), 0,
1503                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1504         files_cachep = kmem_cache_create("files_cache",
1505                         sizeof(struct files_struct), 0,
1506                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1507         fs_cachep = kmem_cache_create("fs_cache",
1508                         sizeof(struct fs_struct), 0,
1509                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1510         mm_cachep = kmem_cache_create("mm_struct",
1511                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1512                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1513         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1514         mmap_init();
1515 }
1516
1517 /*
1518  * Check constraints on flags passed to the unshare system call and
1519  * force unsharing of additional process context as appropriate.
1520  */
1521 static void check_unshare_flags(unsigned long *flags_ptr)
1522 {
1523         /*
1524          * If unsharing a thread from a thread group, must also
1525          * unshare vm.
1526          */
1527         if (*flags_ptr & CLONE_THREAD)
1528                 *flags_ptr |= CLONE_VM;
1529
1530         /*
1531          * If unsharing vm, must also unshare signal handlers.
1532          */
1533         if (*flags_ptr & CLONE_VM)
1534                 *flags_ptr |= CLONE_SIGHAND;
1535
1536         /*
1537          * If unsharing namespace, must also unshare filesystem information.
1538          */
1539         if (*flags_ptr & CLONE_NEWNS)
1540                 *flags_ptr |= CLONE_FS;
1541 }
1542
1543 /*
1544  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1545  */
1546 static int unshare_thread(unsigned long unshare_flags)
1547 {
1548         if (unshare_flags & CLONE_THREAD)
1549                 return -EINVAL;
1550
1551         return 0;
1552 }
1553
1554 /*
1555  * Unshare the filesystem structure if it is being shared
1556  */
1557 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1558 {
1559         struct fs_struct *fs = current->fs;
1560
1561         if (!(unshare_flags & CLONE_FS) || !fs)
1562                 return 0;
1563
1564         /* don't need lock here; in the worst case we'll do useless copy */
1565         if (fs->users == 1)
1566                 return 0;
1567
1568         *new_fsp = copy_fs_struct(fs);
1569         if (!*new_fsp)
1570                 return -ENOMEM;
1571
1572         return 0;
1573 }
1574
1575 /*
1576  * Unsharing of sighand is not supported yet
1577  */
1578 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1579 {
1580         struct sighand_struct *sigh = current->sighand;
1581
1582         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1583                 return -EINVAL;
1584         else
1585                 return 0;
1586 }
1587
1588 /*
1589  * Unshare vm if it is being shared
1590  */
1591 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1592 {
1593         struct mm_struct *mm = current->mm;
1594
1595         if ((unshare_flags & CLONE_VM) &&
1596             (mm && atomic_read(&mm->mm_users) > 1)) {
1597                 return -EINVAL;
1598         }
1599
1600         return 0;
1601 }
1602
1603 /*
1604  * Unshare file descriptor table if it is being shared
1605  */
1606 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1607 {
1608         struct files_struct *fd = current->files;
1609         int error = 0;
1610
1611         if ((unshare_flags & CLONE_FILES) &&
1612             (fd && atomic_read(&fd->count) > 1)) {
1613                 *new_fdp = dup_fd(fd, &error);
1614                 if (!*new_fdp)
1615                         return error;
1616         }
1617
1618         return 0;
1619 }
1620
1621 /*
1622  * unshare allows a process to 'unshare' part of the process
1623  * context which was originally shared using clone.  copy_*
1624  * functions used by do_fork() cannot be used here directly
1625  * because they modify an inactive task_struct that is being
1626  * constructed. Here we are modifying the current, active,
1627  * task_struct.
1628  */
1629 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1630 {
1631         int err = 0;
1632         struct fs_struct *fs, *new_fs = NULL;
1633         struct sighand_struct *new_sigh = NULL;
1634         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1635         struct files_struct *fd, *new_fd = NULL;
1636         struct nsproxy *new_nsproxy = NULL;
1637         int do_sysvsem = 0;
1638
1639         check_unshare_flags(&unshare_flags);
1640
1641         /* Return -EINVAL for all unsupported flags */
1642         err = -EINVAL;
1643         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1644                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1645                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1646                 goto bad_unshare_out;
1647
1648         /*
1649          * CLONE_NEWIPC must also detach from the undolist: after switching
1650          * to a new ipc namespace, the semaphore arrays from the old
1651          * namespace are unreachable.
1652          */
1653         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1654                 do_sysvsem = 1;
1655         if ((err = unshare_thread(unshare_flags)))
1656                 goto bad_unshare_out;
1657         if ((err = unshare_fs(unshare_flags, &new_fs)))
1658                 goto bad_unshare_cleanup_thread;
1659         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1660                 goto bad_unshare_cleanup_fs;
1661         if ((err = unshare_vm(unshare_flags, &new_mm)))
1662                 goto bad_unshare_cleanup_sigh;
1663         if ((err = unshare_fd(unshare_flags, &new_fd)))
1664                 goto bad_unshare_cleanup_vm;
1665         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1666                         new_fs)))
1667                 goto bad_unshare_cleanup_fd;
1668
1669         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1670                 if (do_sysvsem) {
1671                         /*
1672                          * CLONE_SYSVSEM is equivalent to sys_exit().
1673                          */
1674                         exit_sem(current);
1675                 }
1676
1677                 if (new_nsproxy) {
1678                         switch_task_namespaces(current, new_nsproxy);
1679                         new_nsproxy = NULL;
1680                 }
1681
1682                 task_lock(current);
1683
1684                 if (new_fs) {
1685                         fs = current->fs;
1686                         spin_lock(&fs->lock);
1687                         current->fs = new_fs;
1688                         if (--fs->users)
1689                                 new_fs = NULL;
1690                         else
1691                                 new_fs = fs;
1692                         spin_unlock(&fs->lock);
1693                 }
1694
1695                 if (new_mm) {
1696                         mm = current->mm;
1697                         active_mm = current->active_mm;
1698                         current->mm = new_mm;
1699                         current->active_mm = new_mm;
1700                         if (current->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) {
1701                                 atomic_dec(&mm->oom_disable_count);
1702                                 atomic_inc(&new_mm->oom_disable_count);
1703                         }
1704                         activate_mm(active_mm, new_mm);
1705                         new_mm = mm;
1706                 }
1707
1708                 if (new_fd) {
1709                         fd = current->files;
1710                         current->files = new_fd;
1711                         new_fd = fd;
1712                 }
1713
1714                 task_unlock(current);
1715         }
1716
1717         if (new_nsproxy)
1718                 put_nsproxy(new_nsproxy);
1719
1720 bad_unshare_cleanup_fd:
1721         if (new_fd)
1722                 put_files_struct(new_fd);
1723
1724 bad_unshare_cleanup_vm:
1725         if (new_mm)
1726                 mmput(new_mm);
1727
1728 bad_unshare_cleanup_sigh:
1729         if (new_sigh)
1730                 if (atomic_dec_and_test(&new_sigh->count))
1731                         kmem_cache_free(sighand_cachep, new_sigh);
1732
1733 bad_unshare_cleanup_fs:
1734         if (new_fs)
1735                 free_fs_struct(new_fs);
1736
1737 bad_unshare_cleanup_thread:
1738 bad_unshare_out:
1739         return err;
1740 }
1741
1742 /*
1743  *      Helper to unshare the files of the current task.
1744  *      We don't want to expose copy_files internals to
1745  *      the exec layer of the kernel.
1746  */
1747
1748 int unshare_files(struct files_struct **displaced)
1749 {
1750         struct task_struct *task = current;
1751         struct files_struct *copy = NULL;
1752         int error;
1753
1754         error = unshare_fd(CLONE_FILES, &copy);
1755         if (error || !copy) {
1756                 *displaced = NULL;
1757                 return error;
1758         }
1759         *displaced = task->files;
1760         task_lock(task);
1761         task->files = copy;
1762         task_unlock(task);
1763         return 0;
1764 }